feat: qualify compact LOD phase 5

.github/workflows/benchmark.yml

@@ -13,7 +13,12 @@ on:
       duration:
         description: Benchmark duration per preset (seconds)
         required: false
-        default: "5"
+        default: "60"
+      capture_gpu_culling:
+        description: Capture the bounded gpu-culling-scale extreme/traversal CPU-vs-GPU source pair
+        required: false
+        type: boolean
+        default: false
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
@@ -35,7 +40,8 @@ jobs:
       issues: write
       statuses: write
     runs-on: blacksmith-2vcpu-ubuntu-2404
-    timeout-minutes: 30
+    # Long scheduled/manual acceptance captures remain externally bounded.
+    timeout-minutes: 60
     steps:
       - uses: actions/checkout@v4
 
@@ -63,7 +69,7 @@ jobs:
             });
             const shouldRun = files.some((file) => {
               const name = file.filename;
-              return name === 'build.zig' || name.startsWith('src/') || name.startsWith('modules/') || name.startsWith('assets/shaders/');
+              return name === 'build.zig' || name.startsWith('src/') || name.startsWith('modules/') || name.startsWith('assets/shaders/') || name.startsWith('scripts/') || name.startsWith('docs/benchmarks/') || name === '.github/workflows/benchmark.yml';
             });
             core.setOutput('run', shouldRun ? 'true' : 'false');
 
@@ -96,28 +102,96 @@ jobs:
         uses: ./.github/actions/run-with-log
         with:
           name: Benchmark
-          timeout: 20m
+          timeout: ${{ github.event_name == 'schedule' && '55m' || github.event_name == 'workflow_dispatch' && '55m' || '20m' }}
           log-file: benchmark.log
-          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && nix develop .#ci-graphics --command bash scripts/run_benchmark.sh --duration "$BENCHMARK_DURATION" --presets low,medium,high --output-dir benchmark-results --per-preset-timeout 600
+          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && nix develop .#ci-graphics --command bash scripts/run_benchmark.sh --duration "$BENCHMARK_DURATION" --presets low,medium,high --scenarios stationary,traversal,rapid-turn,teleport-eviction --compact-modes off,auto --benchmark-world overworld --output-dir benchmark-results --per-preset-timeout 600
         env:
           XDG_RUNTIME_DIR: /tmp/runtime-runner
           WAYLAND_DISPLAY: headless
           MESA_SHADER_CACHE_DIR: /tmp/mesa_shader_cache
           SDL_AUDIODRIVER: dummy
           ZIGCRAFT_SAFE_MODE: "1"
-          BENCHMARK_DURATION: ${{ github.event_name == 'workflow_dispatch' && github.event.inputs.duration || '5' }}
+          # PR/push runs are bounded canaries. Scheduled and manually requested
+          # captures are longer acceptance evidence, not synthetic fixtures.
+          BENCHMARK_DURATION: ${{ github.event_name == 'schedule' && '60' || github.event_name == 'workflow_dispatch' && github.event.inputs.duration || '5' }}
+          ZIGCRAFT_BENCHMARK_GPU_ADAPTER: Lavapipe (Mesa software Vulkan)
+          ZIGCRAFT_BENCHMARK_GPU_DRIVER: Mesa Lavapipe supplied by the pinned Nix environment
+          ZIGCRAFT_BENCHMARK_RUNNER: blacksmith-2vcpu-ubuntu-2404 GitHub Actions runner
+          ZIGCRAFT_BENCHMARK_ZIG_TOOLCHAIN: Zig 0.16.0 supplied by the pinned Nix flake
+
+      - name: Run Phase 5 long-session streaming stress gate
+        if: steps.gate.outputs.run == 'true'
+        uses: ./.github/actions/run-with-log
+        with:
+          name: Phase 5 Streaming Stress
+          timeout: 15m
+          log-file: phase5-streaming-stress.log
+          command: nix develop .#ci-unit --command zig build phase5-stress-gate -Dphase5-stress-iterations="$PHASE5_STRESS_ITERATIONS"
+        env:
+          # The scheduled run is longer by operation count, never by a flaky
+          # elapsed-time threshold. PR feedback stays bounded.
+          PHASE5_STRESS_ITERATIONS: ${{ github.event_name == 'schedule' && '768' || '64' }}
+
+      - name: Validate compact benchmark evidence matrix
+        if: steps.gate.outputs.run == 'true'
+        run: python3 scripts/benchmark_baseline.py validate-compact-matrix benchmark-results
+
+      - name: Capture dedicated CPU-vs-GPU culling sources
+        # Scheduled captures always retain a fresh pair. Manual runs opt in so
+        # ordinary canaries stay fast; each source is independently bounded.
+        if: steps.gate.outputs.run == 'true' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.capture_gpu_culling == 'true'))
+        uses: ./.github/actions/run-with-log
+        with:
+          name: GPU Culling Baseline Sources
+          timeout: 32m
+          log-file: gpu-culling-benchmark.log
+          command: mkdir -p "$MESA_SHADER_CACHE_DIR" && nix develop .#ci-graphics --command bash -c 'scripts/run_benchmark.sh --duration 60 --presets extreme --scenarios traversal --compact-modes auto --gpu-culling off --gpu-culling-threshold 128 --benchmark-fixture gpu-culling-scale --benchmark-horizon-distance 4096 --benchmark-lod-memory-budget-mb 2048 --benchmark-require-gpu-candidates 1024 --benchmark-world flat --output-dir gpu-culling-results/cpu --per-preset-timeout 900 && scripts/run_benchmark.sh --duration 60 --presets extreme --scenarios traversal --compact-modes auto --gpu-culling on --gpu-culling-threshold 128 --benchmark-fixture gpu-culling-scale --benchmark-horizon-distance 4096 --benchmark-lod-memory-budget-mb 2048 --benchmark-require-gpu-candidates 1024 --benchmark-world flat --output-dir gpu-culling-results/gpu --per-preset-timeout 900'
+        env:
+          XDG_RUNTIME_DIR: /tmp/runtime-runner
+          WAYLAND_DISPLAY: headless
+          MESA_SHADER_CACHE_DIR: /tmp/mesa_shader_cache
+          SDL_AUDIODRIVER: dummy
+          ZIGCRAFT_SAFE_MODE: "1"
+          ZIGCRAFT_BENCHMARK_GPU_ADAPTER: Lavapipe (Mesa software Vulkan)
+          ZIGCRAFT_BENCHMARK_GPU_DRIVER: Mesa Lavapipe supplied by the pinned Nix environment
+          ZIGCRAFT_BENCHMARK_RUNNER: blacksmith-2vcpu-ubuntu-2404 GitHub Actions runner
+          ZIGCRAFT_BENCHMARK_ZIG_TOOLCHAIN: Zig 0.16.0 supplied by the pinned Nix flake
+
+      - name: Validate fresh CPU-vs-GPU culling pair
+        if: steps.gate.outputs.run == 'true' && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && github.event.inputs.capture_gpu_culling == 'true'))
+        run: |
+          python3 scripts/benchmark_baseline.py assemble-gpu-culling --overwrite \
+            --output gpu-culling-results/gpu-culling-baseline.json \
+            gpu-culling-results/cpu/auto/extreme/traversal.json \
+            gpu-culling-results/gpu/auto/extreme/traversal.json
+          python3 scripts/benchmark_baseline.py validate-gpu-culling \
+            gpu-culling-results/gpu-culling-baseline.json
 
       - name: Stop headless Wayland compositor
         if: always() && steps.gate.outputs.run == 'true'
         uses: ./.github/actions/stop-weston
 
-      - name: Compare against baseline
+      - name: Reject Vulkan validation errors in benchmark log
         if: steps.gate.outputs.run == 'true'
+        run: |
+          if rg -n -i 'vuid-|validation.*(error|failed)|(error|failed).*validation' benchmark.log; then
+            printf 'Vulkan validation errors were reported during the benchmark.\n' >&2
+            exit 1
+          fi
+
+      - name: Compare against baseline
+        # A 5-second canary must not be numerically compared with a 60-second
+        # acceptance capture. The compatibility command also enforces this.
+        if: steps.gate.outputs.run == 'true' && github.event_name != 'schedule' && github.event_name != 'workflow_dispatch'
         id: compare
         run: |
-          for preset in low medium high; do
-            bash scripts/compare_benchmarks.sh docs/benchmarks/baseline.json "benchmark-results/${preset}.json" --preset "$preset"
-          done
+          if ! python3 scripts/benchmark_baseline.py compatibility docs/benchmarks/baseline.json benchmark-results/auto/low/stationary.json --preset low --scenario stationary; then
+            printf 'Skipping numeric baseline comparison: checked reference hardware/provenance differs from this CI runner. Matrix, SLO, validation, and Bencher gates still apply.\n'
+            exit 0
+          fi
+          for preset in low medium high; do for scenario in stationary traversal rapid-turn teleport-eviction; do
+            bash scripts/compare_benchmarks.sh docs/benchmarks/baseline.json "benchmark-results/auto/${preset}/${scenario}.json" --preset "$preset" --scenario "$scenario"
+          done; done
 
       - name: Install Bencher CLI
         if: steps.gate.outputs.run == 'true' && env.BENCHER_API_TOKEN != '' && env.BENCHER_PROJECT != ''
@@ -128,16 +202,16 @@ jobs:
         env:
           BENCHER_BRANCH: ${{ github.head_ref || github.ref_name }}
         run: |
-          for preset in low medium high; do
+          for preset in low medium high; do for scenario in stationary traversal rapid-turn teleport-eviction; do
             bencher run \
               --project "$BENCHER_PROJECT" \
               --branch "$BENCHER_BRANCH" \
               --testbed "github-actions-lavapipe" \
               --token "$BENCHER_API_TOKEN" \
               --adapter json \
               --err \
-              "cat benchmark-results/${preset}.json"
-          done
+              "cat benchmark-results/auto/${preset}/${scenario}.json"
+          done; done
 
       - name: Comment Bencher dashboard
         if: steps.gate.outputs.run == 'true' && github.event_name == 'pull_request' && env.BENCHER_DASHBOARD_URL != ''
@@ -158,7 +232,10 @@ jobs:
           name: benchmark-results
           path: |
             benchmark-results
+            gpu-culling-results
             benchmark.log
+            gpu-culling-benchmark.log
+            phase5-streaming-stress.log
             weston.log
           retention-days: 30
 

.github/workflows/build.yml

@@ -9,6 +9,7 @@ on:
       - "libs/**"
       - "assets/shaders/**"
       - "scripts/**"
+      - "docs/benchmarks/**"
       - "build.zig"
       - "build.zig.zon"
       - "flake.nix"
@@ -18,6 +19,7 @@ on:
       - ".github/actions/setup-lavapipe/**"
       - ".github/vulkan/**"
       - ".github/workflows/build.yml"
+      - ".github/workflows/benchmark.yml"
   pull_request:
     branches: [dev]
     paths:
@@ -26,6 +28,7 @@ on:
       - "libs/**"
       - "assets/shaders/**"
       - "scripts/**"
+      - "docs/benchmarks/**"
       - "build.zig"
       - "build.zig.zon"
       - "flake.nix"
@@ -35,6 +38,7 @@ on:
       - ".github/actions/setup-lavapipe/**"
       - ".github/vulkan/**"
       - ".github/workflows/build.yml"
+      - ".github/workflows/benchmark.yml"
   workflow_dispatch:
     inputs:
       ref:
@@ -81,6 +85,7 @@ jobs:
               - 'libs/**'
               - 'assets/shaders/**'
               - 'scripts/**'
+              - 'docs/benchmarks/**'
               - 'build.zig'
               - 'build.zig.zon'
               - 'flake.nix'
@@ -90,6 +95,7 @@ jobs:
               - '.github/actions/setup-lavapipe/**'
               - '.github/vulkan/**'
               - '.github/workflows/build.yml'
+              - '.github/workflows/benchmark.yml'
             platform_changes:
               - 'src/**'
               - 'modules/**'
@@ -191,6 +197,14 @@ jobs:
           log-file: unit-test-${{ matrix.optimize }}.log
           command: nix develop .#ci-unit --command zig build -Doptimize=${{ matrix.optimize }} test
 
+      - name: Run Phase 5 compact LOD gate
+        if: needs.changes.outputs.code_changes == 'true' && matrix.optimize == 'Debug'
+        run: nix develop .#ci-unit --command zig build phase5-gate
+
+      - name: Run bounded Phase 5 streaming stress gate
+        if: needs.changes.outputs.code_changes == 'true' && matrix.optimize == 'Debug'
+        run: nix develop .#ci-unit --command zig build phase5-stress-gate -Dphase5-stress-iterations=64
+
       - name: Upload unit test log
         if: failure()
         uses: actions/upload-artifact@v7
@@ -221,7 +235,7 @@ jobs:
       contents: read
       id-token: write
     runs-on: blacksmith-2vcpu-ubuntu-2404
-    timeout-minutes: 45
+    timeout-minutes: 75
     needs: [changes, unit-test]
     steps:
       - uses: actions/checkout@v4
@@ -274,15 +288,38 @@ jobs:
           ZIGCRAFT_SMOKE_FRAMES: "3"
           ZIGCRAFT_SAFE_MODE: "1"
 
+      - name: Run Phase 5 production visual-motion and saved-world gate
+        if: needs.changes.outputs.code_changes == 'true'
+        uses: ./.github/actions/run-with-log
+        with:
+          name: Phase 5 Visual Motion and Saved-World Reload Gate
+          timeout: 35m
+          log-file: phase5-visual-gate.log
+          command: nix develop .#ci-graphics --command zig build phase5-visual-gate
+        env:
+          XDG_RUNTIME_DIR: /tmp/runtime-runner
+          WAYLAND_DISPLAY: headless
+
+      - name: Upload Phase 5 visual evidence
+        if: always() && needs.changes.outputs.code_changes == 'true'
+        uses: actions/upload-artifact@v7
+        with:
+          name: phase5-visual-evidence
+          path: |
+            zig-out/phase5-visual-smoke/
+            phase5-visual-gate.log
+          if-no-files-found: ignore
+          retention-days: 14
+
       - name: Fail on Vulkan validation log errors
         if: needs.changes.outputs.code_changes == 'true'
         run: |
           set -euo pipefail
-          if grep -E "Vulkan validation error|Validation Error:|VUID-" integration-test.log world-smoke-test.log; then
-            echo "Vulkan validation errors were found in integration logs." >&2
+          if rg -n -e 'Vulkan validation error|Validation Error:|VUID-' integration-test.log world-smoke-test.log phase5-visual-gate.log; then
+            echo "Vulkan validation errors were found in integration or Phase 5 visual logs." >&2
             exit 1
           fi
-          echo "No Vulkan validation errors found in integration logs."
+          echo "No Vulkan validation errors found in integration or Phase 5 visual logs."
 
       - name: Stop headless Wayland compositor
         if: always() && needs.changes.outputs.code_changes == 'true'
@@ -296,6 +333,7 @@ jobs:
           path: |
             integration-test.log
             world-smoke-test.log
+            phase5-visual-gate.log
             weston.log
           if-no-files-found: ignore
           retention-days: 7

.github/workflows/opencode-pr.yml

@@ -82,6 +82,7 @@ jobs:
           ZHIPU_API_KEY: ${{ secrets.ZHIPU_API_KEY }}
         with:
           model: zhipuai-coding-plan/glm-5.2
+          variant: max
           use_github_token: true
           prompt: ${{ env.PROMPT }}
 

.gitignore

@@ -6,6 +6,7 @@ blocks-temp/
 libs/zig-image/
 result
 node_modules/
+__pycache__/
 .direnv/
 *-profile*
 test_output.txt